Automatic merge from submit-queue
Fix bug in status manager TerminatePod
In TerminatePod, we previously pass pod.Status to updateStatusInternal. This is a bug, since it is the original status that we are given. Not only does it skip updates made to container statuses, but in some cases it reverted the pod's status to an earlier version, since it was being passed a stale status initially.
This was the case in #40239 and #41095. As shown in #40239, the pod's status is set to running after it is set to failed, occasionally causing very long delays in pod deletion since we have to wait for this to be corrected.
This PR fixes the bug, adds some helpful debugging statements, and adds a unit test for TerminatePod (which for some reason didnt exist before?).
@kubernetes/sig-node-bugs @vish @Random-Liu
Automatic merge from submit-queue
Make EnableCRI default to true
This change makes kubelet to use the CRI implementation by default,
unless the users opt out explicitly by using --enable-cri=false.
For the rkt integration, the --enable-cri flag will have no effect
since rktnetes does not use CRI.
Also, mark the original --experimental-cri flag hidden and deprecated,
so that we can remove it in the next release. If both flags are specified,
the --enable-cri flag overrides the --experimental-cri flag.
Automatic merge from submit-queue
fix comment
**What this PR does / why we need it**:
fix comment
Thanks.
**Special notes for your reviewer**:
**Release note**:
```release-note
```
This change makes kubelet to use the CRI implementation by default,
unless the users opt out explicitly by using --enable-cri=false.
For the rkt integration, the --enable-cri flag will have no effect
since rktnetes does not use CRI.
Also, mark the original --experimental-cri flag hidden and deprecated,
so that we can remove it in the next release.
To safely mark a volume detached when the volume controller manager is used.
An example of one such problem:
1. pod is created, volume is added to desired state of the world
2. reconciler process starts
3. reconciler starts MountVolume, which is kicked off asynchronously via
operation_executor.go
4. MountVolume mounts the volume, but hasn't yet marked it as mounted
5. pod is deleted, volume is removed from desired state of the world
6. reconciler detects volume is no longer in desired state of world,
removes it from volumes in use
7. MountVolume tries to mark volume in use, throws an error because
volume is no longer in actual state of world list.
8. controller-manager tries to detach the volume, this fails because it
is still mounted to the OS.
9. EBS gets stuck indefinitely in busy state trying to detach.
Automatic merge from submit-queue (batch tested with PRs 40796, 40878, 36033, 40838, 41210)
Implement TTL controller and use the ttl annotation attached to node in secret manager
For every secret attached to a pod as volume, Kubelet is trying to refresh it every sync period. Currently Kubelet has a ttl-cache of secrets of its pods and the ttl is set to 1 minute. That means that in large clusters we are targetting (5k nodes, 30pods/node), given that each pod has a secret associated with ServiceAccount from its namespaces, and with large enough number of namespaces (where on each node (almost) every pod is from a different namespace), that resource in ~30 GETs to refresh all secrets every minute from one node, which gives ~2500QPS for GET secrets to apiserver.
Apiserver cannot keep up with it very easily.
Desired solution would be to watch for secret changes, but because of security we don't want a node watching for all secrets, and it is not possible for now to watch only for secrets attached to pods from my node.
So as a temporary solution, we are introducing an annotation that would be a suggestion for kubelet for the TTL of secrets in the cache and a very simple controller that would be setting this annotation based on the cluster size (the large cluster is, the bigger ttl is).
That workaround mean that only very local changes are needed in Kubelet, we are creating a well separated very simple controller, and once watching "my secrets" will be possible it will be easy to remove it and switch to that. And it will allow us to reach scalability goals.
@dchen1107 @thockin @liggitt
Automatic merge from submit-queue (batch tested with PRs 41074, 41147, 40854, 41167, 40045)
Add debug logging to eviction manager
**What this PR does / why we need it**:
This PR adds debug logging to eviction manager.
We need it to help users understand when/why eviction manager is/is not making decisions to support information gathering during support.
Automatic merge from submit-queue (batch tested with PRs 40873, 40948, 39580, 41065, 40815)
[CRI] Enable Hostport Feature for Dockershim
Commits:
1. Refactor common hostport util logics and add more tests
2. Add HostportManager which can ADD/DEL hostports instead of a complete sync.
3. Add Interface for retreiving portMappings information of a pod in Network Host interface.
Implement GetPodPortMappings interface in dockerService.
4. Teach kubenet to use HostportManager
Automatic merge from submit-queue (batch tested with PRs 38796, 40823, 40756, 41083, 41105)
Let ReadLogs return when there is a read error.
Fixes a bug in kuberuntime log.
Today, @yujuhong found that once we cancel `kubectl logs -f` with `Ctrl+C`, kuberuntime will keep complaining:
```
27939 kuberuntime_logs.go:192] Failed with err write tcp 10.240.0.4:10250->10.240.0.2:53913: write: broken pipe when writing log for log file "/var/log/pods/5bb76510-ed71-11e6-ad02-42010af00002/busybox_0.log": &{timestamp:{sec:63622095387 nsec:625309193 loc:0x484c440} stream:stdout log:[84 117 101 32 70 101 98 32 32 55 32 50 48 58 49 54 58 50 55 32 85 84 67 32 50 48 49 55 10]}
```
This is because kuberuntime keeps writing to the connection even though it is already closed. Actually, kuberuntime should return and report error whenever there is a writing error.
Ref the [docker code](3a4ae1f661/pkg/stdcopy/stdcopy.go (L159-L167))
I'm still creating the cluster and verifying this fix. Will post the result here after that.
/cc @yujuhong @kubernetes/sig-node-bugs
Automatic merge from submit-queue (batch tested with PRs 38796, 40823, 40756, 41083, 41105)
kubelet/network-cni-plugin: modify the log's info
**What this PR does / why we need it**:
Checking the startup logs of kubelet, i can always find a error like this:
"E1215 10:19:24.891724 2752 cni.go:163] error updating cni config: No networks found in /etc/cni/net.d"
It will appears, neither i use cni network-plugin or not.
After analysis codes, i thought it should be a warn log, because it will not produce any actions like as exit or abort, and just ignored when not any valid plugins exit.
thank you!
Automatic merge from submit-queue (batch tested with PRs 41103, 41042, 41097, 40946, 40770)
Use Clientset interface in KubeletDeps
**What this PR does / why we need it**:
This replaces the Clientset struct with the equivalent interface for the KubeClient injected via KubeletDeps. This is useful for testing and for accessing the Node and Pod status event stream without an API server.
**Special notes for your reviewer**:
Follow up to #4907
**Release note**:
`NONE`
Automatic merge from submit-queue (batch tested with PRs 41103, 41042, 41097, 40946, 40770)
dockershim: set security option separators based on the docker version
Also add a version cache to avoid hitting the docker daemon frequently.
This is part of #38164
Automatic merge from submit-queue (batch tested with PRs 40971, 41027, 40709, 40903, 39369)
Set docker opt separator correctly for SELinux options
This is based on @pmorie's commit from #40179
Automatic merge from submit-queue (batch tested with PRs 40930, 40951)
Fix CRI port forwarding
Websocket support was introduced #33684, which broke the CRI
implementation. This change fixes it.
Automatic merge from submit-queue (batch tested with PRs 40289, 40877, 40879, 39972, 40942)
Rename experimental-cgroups-per-pod flag
**What this PR does / why we need it**:
1. Rename `experimental-cgroups-per-qos` to `cgroups-per-qos`
1. Update hack/local-up-cluster to match `CGROUP_DRIVER` with docker runtime if used.
**Special notes for your reviewer**:
We plan to roll this feature out in the upcoming release. Previous node e2e runs were running with this feature on by default. We will default this feature on for all e2es next week.
**Release note**:
```release-note
Rename --experiemental-cgroups-per-qos to --cgroups-per-qos
```
Automatic merge from submit-queue
CRI: Handle cri in-place upgrade
Fixes https://github.com/kubernetes/kubernetes/issues/40051.
## How does this PR restart/remove legacy containers/sandboxes?
With this PR, dockershim will convert and return legacy containers and infra containers as regular containers/sandboxes. Then we can rely on the SyncPod logic to stop the legacy containers/sandboxes, and the garbage collector to remove the legacy containers/sandboxes.
To forcibly trigger restart:
* For infra containers, we manually set `hostNetwork` to opposite value to trigger a restart (See [here](https://github.com/kubernetes/kubernetes/blob/master/pkg/kubelet/kuberuntime/kuberuntime_manager.go#L389))
* For application containers, they will be restarted with the infra container.
## How does this PR avoid extra overhead when there is no legacy container/sandbox?
For the lack of some labels, listing legacy containers needs extra `docker ps`. We should not introduce constant performance regression for legacy container cleanup. So we added the `legacyCleanupFlag`:
* In `ListContainers` and `ListPodSandbox`, only do extra `ListLegacyContainers` and `ListLegacyPodSandbox` when `legacyCleanupFlag` is `NotDone`.
* When dockershim starts, it will check whether there are legacy containers/sandboxes.
* If there are none, it will mark `legacyCleanupFlag` as `Done`.
* If there are any, it will leave `legacyCleanupFlag` as `NotDone`, and start a goroutine periodically check whether legacy cleanup is done.
This makes sure that there is overhead only when there are legacy containers/sandboxes not cleaned up yet.
## Caveats
* In-place upgrade will cause kubelet to restart all running containers.
* RestartNever container will not be restarted.
* Garbage collector sometimes keep the legacy containers for a long time if there aren't too many containers on the node. In that case, dockershim will keep performing extra `docker ps` which introduces overhead.
* Manually remove all legacy containers will fix this.
* Should we garbage collect legacy containers/sandboxes in dockershim by ourselves? /cc @yujuhong
* Host port will not be reclaimed for the lack of checkpoint for legacy sandboxes. https://github.com/kubernetes/kubernetes/pull/39903 /cc @freehan
/cc @yujuhong @feiskyer @dchen1107 @kubernetes/sig-node-api-reviews
**Release note**:
```release-note
We should mention the caveats of in-place upgrade in release note.
```
Automatic merge from submit-queue
Optionally avoid evicting critical pods in kubelet
For #40573
```release-note
When feature gate "ExperimentalCriticalPodAnnotation" is set, Kubelet will avoid evicting pods in "kube-system" namespace that contains a special annotation - `scheduler.alpha.kubernetes.io/critical-pod`
This feature should be used in conjunction with the rescheduler to guarantee availability for critical system pods - https://kubernetes.io/docs/admin/rescheduler/
```
Automatic merge from submit-queue (batch tested with PRs 40795, 40863)
Use caching secret manager in kubelet
I just found that this is in my local branch I'm using for testing, but not in master :)
